Get started
Five minutes, start to finish. Pick what you want to do.
-
Sign in
Loading...
Sign in with Google to get a free API key. Community is free because you pay with data instead of money: conversations are used for training only with your explicit consent. See the Community data policy.
Signed in as .
-
Create an API key
Copy this key now. It will not be shown again. It stays in this tab only. If you reload, create another one.
-
Pick your tool
You do not have a key in this tab, so the snippets show
YOUR_API_KEY. Create a key in step 02 to fill it in.Put your key in the environment, then save this as
opencode.jsonin your project (or~/.config/opencode/opencode.json).Then run
opencodeand picktessovix/tessovix-miniwith/models.In Cline, open Settings and set these fields.
Add this model to
~/.continue/config.yaml(or the Continue config for your workspace).Install the SDK with
pip install openai. -
Send your first request
Use your tool, or run this from a terminal. We will notice the request on our side.
-
Verified
Waiting for your first request...
We saw your request. You are connected.
-
Install the node
Linux (x86_64, aarch64) or macOS (arm64). Run this in a terminal.
You do not have a key in this tab, so the command shows
YOUR_API_KEY. Create a key in step 02 to fill it in.What this does
- Downloads
tessovix-nodefor your platform from beaver.tessovix.com and checks its SHA-256. It stops if the checksum does not match. - Installs the binary to
~/.local/bin. - Saves your key and the gateway address in
~/.config/tessovix/node.env, readable only by you. - On Linux with systemd, creates and starts a
tessovix-nodeuser service. - Does not install an inference engine or models. That is the next part.
Manual alternative
Download
tessovix-node-<target>.tar.gzand its.sha256fromhttps://beaver.tessovix.com/dl/, compare the checksum, and unpacktessovix-nodeinto~/.local/bin. Then run it with these variables set:Start an engine
The node serves requests from a local llama.cpp server on
127.0.0.1:8081. Use the model file and alias from your beta invitation.Get a CUDA, Metal or Vulkan build of llama.cpp, then:
llama-server -m ~/models/MODEL.gguf --alias ALIAS \ --jinja -ngl 99 -c 16384 \ --host 127.0.0.1 --port 8081No GPU is fine for small models only (about 3B parameters or less, 4-bit). Bigger models answer too slowly on a CPU.
llama-server -m ~/models/MODEL.gguf --alias ALIAS \ --jinja -c 8192 -t "$(nproc)" \ --host 127.0.0.1 --port 8081 - Downloads
-
Wait for your node
Waiting for a node to connect...
-
Done
Your node is online and verified. It now serves requests.
Still nothing after 5 minutes. Check the steps above, then